MPRAMunich Personal RePEc Archive

Driven to distraction: Extraneous eventsand underreaction to earnings news

David Hirshleifer and Sonya Seongyeon Lim and Siew Hong

Teoh

Merage School of Business, University of California, Irvine

15. March 2006

Online at http://mpra.ub.uni-muenchen.de/3110/MPRA Paper No. 3110, posted 7. May 2007

April 16, 2007

Driven to Distraction:Extraneous Events

and Underreaction to Earnings News

David Hirshleifer∗

Sonya Seongyeon Lim∗∗

Siew Hong Teoh∗

∗ The Paul Merage School of Business, University of California, Irvine, CA 92697-3125;Hirshleifer: david.h@uci.edu, (949) 824-9955; Teoh: steoh@uci.edu, (949) 824-9952

∗∗ The Kellstadt Graduate School of Business, DePaul University, 1 E. Jackson Blvd,Chicago, IL 60604; slim1@depaul.edu, (312) 362-8825

We thank Nick Barberis, Werner DeBondt, Stefano DellaVigna (NBER conference dis-cussant), Laura Field, Wayne Guay (FRA conference discussant), Christo Karuna, ErikLie, Ray Pfeiffer (FARS conference discussant), Mort Pincus, Charles Shi, seminar par-ticipants at the Merage School of Business at UC Irvine; the Kellstadt Graduate Schoolof Business at DePaul University; the Anderson Graduate School of Management atUCLA; the Sauder School of Business at University of British Columbia; and confer-ence participants at the 10th Biennial Behavioral Decision Research in ManagementConference at Santa Monica, California; the NBER Behavioral Finance November 2006Meeting at Cambridge, Massachusetts; the Financial Research Association 2006 Confer-ence at Las Vegas, Nevada; and the Financial Accounting and Reporting Section 2007Conference at San Antonio, Texas for very helpful comments.

Driven to Distraction:Extraneous Events

and Underreaction to Earnings News

Abstract

Psychological evidence indicates that it is hard to process multiple stimuli and per-form multiple tasks at the same time. This paper tests the investor distraction hypothe-sis, which holds that the arrival of extraneous news causes trading and market prices toreact sluggishly to relevant news about a firm. Our test focuses on the competition forinvestor attention between a firm’s earnings announcements and the earnings announce-ments of other firms. We find that the immediate stock price and volume reaction toa firm’s earnings surprise is weaker, and post-earnings announcement drift is stronger,when a greater number of earnings announcements by other firms are made on the sameday. Distracting news has a stronger effect on firms that receive positive than negativeearnings surprises. Industry-unrelated news has a stronger distracting effect than relatednews. A trading strategy that exploits post-earnings announcement drift is unprofitablefor announcements made on days with little competing news.

[Attention] is the taking possession by the mind in clear and vivid form, ofone out of what seem several simultaneously possible objects or trains ofthought...It implies withdrawal from some things in order to deal effectivelywith others.

William James, Principles of Psychology, 1890

Almost a quarter of British motorists admit they have been so distracted byroadside billboards of semi-naked models that they have dangerously veeredout of their lanes.

Reuters (London), November 21, 2005

1 Introduction

Since minds are finite, attention must be allocated selectively. When individuals try

to process multiple information sources or perform multiple tasks simultaneously, per-

formance suffers.1 Indeed, conscious thought requires a focus on particular ideas or

information to the exclusion of others.

These elemental facts suggest that limited attention affects the perceptions and be-

havior of investors. Specifically, an investor’s effort to understand the implications of

a news announcement by one firm for that firm may interfere with the processing of

information about another firm arriving at the same time. Although there is recent

empirical research on the effects of limited investor attention on securities prices, this

basic prediction has not to our knowledge been tested.

A recent theoretical literature models how constraints on processing multiple infor-

mation signals affects beliefs perceptions and security market prices.2 These models

imply that investor neglect of information signals can lead to serial correlation in as-

set return volatility (Peng, Xiong, and Bollerslev (2006)), mispricing that is related to

publicly available accounting information (Hirshleifer and Teoh (2003)), excessive asset

price comovement (Peng and Xiong (2006)), faster rate of incorporation of information

by large than by small stocks (Peng (2005)), and neglect of long-term public information

(DellaVigna and Pollet (2007)). There has also been analysis of how firms can exploit

limited investor attention by disclosing bad news at times when other firms are making

1See, e.g., Kahneman (1973), Riley and Roitblat (1978), Pashler (1998); Baddeley (1990) reviewsinterference effects in the recall of stored information.

2Research that examines the effects of limited attention on individual decisions such as tradinginclude Sims (2003), Gabaix, Laibson, Moloche, and Weinberg (2006), and Gabaix and Laibson (2004).

1

salient disclosures (Hirshleifer, Lim, and Teoh (2004)), or on days of the week when

investors are less attentive (DellaVigna and Pollet (2006)).

In the models of DellaVigna and Pollet (2006) and Hirshleifer and Teoh (2005), in-

vestors are risk averse, and a subset neglect the information contained in a firm’s latest

earnings realization about future profitability. In equilibrium stock prices underreact to

earnings surprises, so that prices are on average too low after favorable surprises and

too high after unfavorable surprises. In consequence, positive surprises predict high sub-

sequent returns and negative surprises predict low subsequent returns. In other words,

there is post-earnings announcement drift, as documented by Bernard and Thomas

(1989).

A further empirical implication of these models is that when the amount of attention

investors direct toward a firm decreases, there should be more severe underreaction to

its earnings surprises, intensifying subsequent drift. We argue here that the amount of

attention toward a given firm is likely to be smaller when there are more extraneous

news events distracting investors from that firm.3 Therefore, greater distraction implies

more severe underreaction to the firm’s earnings news – a weaker immediate reaction

to the earnings surprise, and stronger post-earnings announcement drift. Intuitively, we

also expect that the greater the distraction, the weaker the trading volume response to

a news announcement.

Together, we call these predictions the investor distraction hypothesis. In this paper

we test the investor distraction hypothesis by measuring when a greater number of

public disclosures by other firms compete for investor attention; and whether greater

distraction reduces volume of trade.

For at least two reasons, the stock market’s processing of a firm’s earnings announce-

ments provides an attractive test of whether investors are able to filter away extraneous

news. First, earnings announcements are frequent, quantifiable, and directly value-

relevant. Second, several pieces of evidence suggest that limited attention affects stock

price reactions to a firm’s earnings announcements. The post-earnings announcement

drift anomaly (Bernard and Thomas (1989)) suggests that some investors at least tem-

porarily neglect the information in earnings surprises about future profitability. Fur-

thermore, there is evidence that market reactions to earnings announcements are more

prompt and complete when there is reason to think investors are paying attention to

earnings: during trading hours rather than non-trading hours (Francis, Pagach, and

Stephan (1992), Bagnoli, Clement, and Watts (2005)), on non-Friday weekdays rather

3Distracting news may also create information processing bottlenecks for analysts. For example, aninternal review committee’s job of evaluating analyst reports is harder when there is more informationto process.

2

than on Fridays as the weekend approaches (DellaVigna and Pollet (2006)), and during

up markets rather than down markets (Hou, Peng, and Xiong (2006)).

The competing news events that we examine are also earnings surprises. Since all

publicly traded U.S. firms need to make earnings announcements, earnings surprises

provide an extensive sample of distracting events. Of course, earnings announcements

by other firms can be relevant for the value of a given test firm. Indeed, there is literature

that tests whether earnings announcements conveys information across firms. However,

even if such information transfer exists, each firm’s earnings announcement is typically

much more informative about its own value than about the value of other firms. Thus,

if attention is limited, earnings announcements by other firms can call investor attention

away from the purpose of valuing the given firm. Such distraction by extraneous news

can potentially weaken the market reaction to its earnings surprise.

Our study adds to a recent literature that provides evidence suggesting that lim-

ited attention may affect both market prices and the decisions of investors and financial

professionals. Evidence that stock prices underreact to public news events4, and that

information seems to diffuse gradually across industries, between large and small firms,

between economically linked firms, and between firms that are followed by different num-

bers of analysts5 is consistent with limited attention causing investors to neglect public

information (although other possible explanations have also been offered). Evidence that

the stock market sometimes reacts to previously-published news6 suggests that relevant

information is neglected at the time of the previous news. Some studies test for the

effects of limited attention by examining how investors trade in response to public news

arrival.7

The past empirical literature on investor attention discussed above has primarily fo-

cused on the neglect of public information signals, and on how greater publicity draws

4On the new issues puzzle, see Loughran and Ritter (1995); on the repurchase anomaly, see Ikenberry,Lakonishok, and Vermaelen (1995); on other types of events, see the review of Hirshleifer (2001). Recentpapers also test whether investors neglect demographic information (DellaVigna and Pollet (2007)) andinformation in oil prices (Pollet (2005)). Klibanoff, Lamont, and Wizman (1998) find that in typicalweeks closed-end country fund prices underreact to shifts in net asset value (NAV), but underreactmuch less during weeks in which news about the country appears on the front page of the New YorkTimes. They argue that this news is redundant given NAV (which is publicly observable), and thereforesuggest that publicity about the country causes the greater reaction in the fund price.

5See, e.g., Brennan, Jegadeesh, and Swaminathan (1993), Cohen and Frazzini (2006), Hong, Lim,and Stein (2000), Hong, Torous, and Valkanov (2007), Hou (2007); Hou and Moskowitz (2005) reportthat delay-prone firms have anomalous returns.

6Ho and Michaely (1988); Huberman and Regev (2001) analyze in detail a case of a particularcompany in which salient reporting of already-public information in the news media about a companyled to extreme price reactions.

7See, e.g., Barber and Odean (2006), Hirshleifer, Myers, Myers, and Teoh (2003), Linnainmaa (2007),and Seasholes and Wu (2005).

3

attention to the firm. A distinctive feature of our paper is that it focuses on the com-

peting signals that draw investor attention away from a given firm. In other words, our

aim is to test directly whether extraneous news distracts investors, causing market prices

to underreact to relevant news.

For our initial tests of the investor distraction hypothesis, we perform quarterly two-

way independent sorts of stocks based on the earnings surprise, and by the number

of earnings announcements by other firms on the same day as the firm’s earnings an-

nouncement. We call days with a large number of competing announcements “high-news

days,” as opposed to “low-news days.” We find that investors’ announcement date re-

actions to earnings news are significantly less sensitive to earnings news on high-news

days than on low-news days; the interdecile spread of announcement-period abnormal

returns between firms with high and low earnings surprises is 7.07% for low-news days

(the bottom number of announcements decile) and 5.67% for high-news days (the top

number of announcements decile).

To ensure that our results are not driven by seasonalities or cross-sectional differ-

ences in stock price reactions to earnings news, we also perform multivariate regressions

that control for the calendar effects (day of week, month, and year) on the sensitivity of

market reactions to earnings surprises, as well as the effect of size, book-to-market, the

number of analysts following, reporting lags, and institutional ownership. Using multi-

variate analysis, we find a similar result that the announcement date return response is

significantly less sensitive to earnings news when there is a greater number of competing

announcements on the same day.8

If competing news is distracting, it may also weaken the trading volume response to

earnings announcements. Using regression analysis, we find that the abnormal trading

volume response to earnings is significantly weakened when the earnings announcement

occurs on a high-news day than on a low-news day. Together, these announcement-

period volume and return findings are consistent with competing news events drawing

investor attention away from a firm’s earnings announcements, and thereby weakening

the initial price and trading response to these announcements.

To further test the investor distraction hypothesis, we examine whether post-earnings

announcement drift is stronger when earnings announcements occur on days with many

competing announcements. When we sort stocks based upon the earnings surprise and

by the number of earnings announcements on the same day, we find that the post-

earnings announcement drift is significantly stronger on high news days. For high-

8In the previous version, we used the residual number of announcements after regressing the numberof announcements on day of week, month, and year dummy variables as a measure of distraction andfound similar results.

4

news days, the interdecile spread of the post-announcement 60-day cumulative abnormal

returns between high earnings surprise firms and low earnings surprise firms is 7.41% and

significant at the 1% level. The spread is only 2.81% and marginally significant at the

10% level for low-news days. Regression analyses also confirm that post-announcement

drift is significantly stronger for earnings announcements made on days with a greater

number of competing announcements after controlling for other possible determinants

of drift.

Taken together, the univariate and multivariate findings that high-news days are

associated with a lower sensitivity of announcement abnormal returns to earnings news,

with a higher sensitivity of post-announcement abnormal returns to earnings news, and

with a lower trading volume response to earnings news, support the investor distraction

hypothesis.

When we examine market reactions to positive and negative earnings news separately,

we find that the distraction effect is greater in firms receiving positive earnings news. A

possible explanation lies in the fact that attention-drawing events including extreme news

for a given firm are on average associated with individual investor purchases (Barber

and Odean (2006)). After negative news, this amounts to contrarian trading, which

tends to mute the market response to the surprise, whereas after positive news, this is

positive feedback trading, which tends to strengthen the immediate market response to

the surprise.

There may be reasons other than distraction why the number of competing announce-

ments affects the sensitivity of returns to earnings. As discussed in Section 4, it is not

entirely obvious why this should be the case. One possible reason is that the number

of competing announcements affects the informativeness of a given firm’s earnings sur-

prise. However, the distraction hypothesis implies that the number of competing news

announcements has opposite effects on the immediate reactivity of the firm’s stock to

its earnings surprise, versus the post-event reactivity. To compete with the distraction

hypothesis, any alternative explanation faces the hurdle of explaining these opposite

effects.

Our findings also suggest that an investor who seeks to exploit post-earnings an-

nouncement drift can gain by taking into account the amount of competing news on

earnings announcement dates. To test whether the number of competing news an-

nouncements is useful information for trading strategies, we form portfolios based upon

earnings surprises and upon the number of competing events on the day of earnings

announcement. At the end of each month, we perform an independent double sort of

stocks into 5 × 5 = 25 groups based on their most recent quarterly earnings announce-

ments within the preceding three months and the number of announcements on the

5

announcement day. The portfolio that is long in good earnings news firms and short in

bad earnings news firms within the lowest number of announcements quintile is called

the “low-news portfolio.” Similarly, the long-short earnings portfolio formed from firms

with the highest number of announcements is termed the “high-news portfolio.”

We find that taking into account extraneous news is useful for portfolio trading

strategies. The Fama-French three-factor alphas associated with these portfolios differ

significantly across the number of announcements quintiles; the 3-factor alpha is 1.64%

per month and highly significant for the high-news portfolio, while it is 0.77% and

insignificant for the low-news portfolio.

Different competing announcements may affect attention differently. Investors who

follow only a particular sector may find earnings announcements by same-industry firms

more distracting than announcements by firms in other industries. On the other hand,

the opposite effect is also possible that announcements by same-industry firms help

investors react more efficiently to a given announcement. For example, the announce-

ment of record-high earnings by Pixar might attract more investor attention to other

entertainment firms’ earnings announcements on the same day, either because of greater

press coverage of the entertainment industry on that day, because the announcement

inherently draws investor attention to the entertainment industry, or because a focus

on one entertainment firm makes it easier for investors to process information about

entertainment industry firms. If such attention-drawing effects occur, the news about

Pixar should then distract investors from earnings announcements by firms in different

industries.

This suggests that there may be a difference between the effect of industry-related

versus industry-unrelated announcements on the attention that investors devote to the

earnings announcements of a given firm. To test for such a difference, we separate

competing announcements into related and unrelated announcements, using the Fama-

French 10 industry classification to measure industry relatedness. We find that a greater

number of unrelated announcements reduces the sensitivity of announcement-period re-

turns to earnings news, consistent with the results using the total number of announce-

ments. Furthermore, this distraction effect is stronger for unrelated announcements

than related announcements. Similarly, a greater number of unrelated announcements

significantly strengthens post-earnings announcement drift; and does so more strongly

than does related news.

Overall, our evidence about announcement period returns, post-earnings announce-

ment drift, and trading volume responses are generally consistent with the investor

distraction hypothesis. These findings therefore suggest that limited investor attention

affects investor behavior and capital market prices.

6

2 Psychological Basis for Distraction Effects

Psychologists have provided a great deal of evidence that it is hard to process multiple

information sources or perform multiple tasks at the same time. The interfering effect

of extraneous information is illustrated by the famous Stroop task (Stroop (1935)), in

which subjects are asked to name the color in which a word is printed, when the word

does not match its print color, e.g., the word “blue” printed in red ink. When the

meaning of the word differs from its print color, subjects are slower to name its color,

as compared, e.g., with naming the color of a geometrical figure.

Selective attention involves the focus (conscious or otherwise) on a portion of a scene

or set of stimuli. In some studies of selective attention, individuals are asked to di-

rect their attention toward a stimulus, which interferes with the processing of another.

In studies of dichotic listening (Cherry (1953), Moray (1959), Broadbent (1958)), two

messages are separately and simultaneously played into a subject’s left and right ear

using headphones. In some studies, subjects are asked to attend to one of two messages,

and ‘shadow’ (repeat back) the words of this message. They are then asked questions

about the message they were not attending to. Subjects absorb very little information

about the unattended message—whether the voice was male or female, but not what

language was spoken or any of the words that were spoken, even if the same word is

spoken repeatedly

In visual studies of selective attention, participants often think that they have ab-

sorbed a scene fully when in fact they have only absorbed the subset of details upon which

they have focused. Selective attention leads to ‘change blindness’ (wherein a noteworthy

change in a visual scene is not noticed; see Simons and Levin (1997)). The phenomenon

of ‘inattentional blindness’ involves the failure to perceive task-unrelated stimuli while

performing a visual observation task. In such experiments, participants often fail to

notice even seemingly conspicuous events in the video scene they are observing—such

as a woman walking by in a gorilla suit, stopping, and beating her chest before moving

on (Simons and Chabris (1999)).

Studies of divided attention and dual task performance ask participants to attend to

multiple stimuli at the same time and to respond to them. In the auditory domain, a

dichotic listening experiment can be used to examine the effects of divided attention. In

such an experiment, subjects can be asked to pay attention to both messages, and later

can be asked about the content of each. Studies of dual task performance have found

that there is interference between tasks (see, e.g., Pashler and Johnston (1998)), and

that performance is much worse when the two tasks are similar, as with tasks involving

the same sensory modalities (McLeod (1977), Treisman and Davies (1973)).

7

In a financial context, the problem of reacting to multiple earnings surprises by

revaluing two different stocks divides attention, and therefore may also be hard to do.

Performing valuations involves using similar kinds of information and types of cognitive

processing, potentially leading to interference between tasks. Regardless of whether this

is the case, more generally, time and cognitive constraints compel restricting attention

to a limited set of inputs and tasks.

An investor who tries to forecast firms’ prospects are faced with the arrival of many

information signals over time. Psychologists have studied experimentally how subjects

learn over time to forecast a variable that is stochastically related to multiple cues.9 A

consistent finding in both animal and human studies is that cue competition occurs: the

arrival of irrelevant cues causes subjects to use relevant cues less. In financial markets,

investors presumably try to economize on attention by filtering away irrelevant signals.

Nevertheless, psychological evidence of cue competition suggests that stock investors

may be more prone to underreact to relevant information about a firm when there is

greater arrival of irrelevant signals.

3 The Data

We use quarterly earnings announcement data from CRSP-Compustat merged database

and IBES from 1995 to 2004. To calculate the daily number of quarterly earnings

announcements, we look at quarterly earnings announcements available from CRSP-

Compustat merged database. When the announcement date is also available at IBES but

is different from Compustat date, we take the earlier date following the imputation rule

of DellaVigna and Pollet (2006).10 While the accuracy of the announcement date is likely

to be higher when it is available from both IBES and Compustat, we include Compustat

earnings announcements without matching IBES data when we compute the number

of competing announcements each day because IBES coverage is for relatively large

firms.11 Our sample firms are limited to those that have IBES coverage; we therefore

expect very accurate announcement dates for our sample even though the number of

competing announcements can be slightly noisy.

To estimate the forecast error (FE) as a measure of the earnings surprise, we cal-

culate the difference between announced earnings as reported by IBES (eiq) and the

consensus earnings forecast (Fiq), defined as the median of the most recent forecasts

9See, e.g., Baker, Mercier, Valleettourangeau, Frank, and Pan (1993), Busemeyer, Myung, andMcDaniel (1993), and Kruschke and Johansen (1999).

10DellaVigna and Pollet (2006) report that the accuracy of announcement dates imputed from IBESand Compustat are almost perfect in the post-1994 period.

11According to Hong, Lim, and Stein (2000), 40% of CRSP firms are not covered by IBES in 1994.

8

from individual analysts. To exclude stale forecasts when we calculate the consensus

forecast, we only include 1- or 2-quarter ahead forecasts issued or reviewed in the last

60 calendar days before the earnings announcement. If an analyst made multiple fore-

casts during that period, we take her most recent forecast. The difference between the

announced earnings and the consensus forecast is normalized by the stock price at the

end of the corresponding quarter (Piq), where earnings, forecasts, and stock prices are all

split-adjusted. To control for possible data errors, we delete observations when earnings

or forecasts are greater than the stock price, or when the stock price is less than $1

before split-adjustment.

FEiq =eiq − FiqPiq

. (1)

The cumulative abnormal returns of the announcement window and the post-announcement

window are defined as the difference between the buy-and-hold return of the announcing

firm and that of a size and book-to-market (B/M) matching portfolio over the windows

[0, 1] and [2, 61] in trading days relative to the announcement date,

CAR[0, 1]iq =t+1∏

k=t

(1 +Rik)−t+1∏

k=t

(1 +Rpk)

CAR[2, 61]iq =t+61∏

k=t+2

(1 +Rik)−t+61∏

k=t+2

(1 +Rpk), (2)

where Rik is the return of the firm i and Rpk is the return of the matching size-B/M

portfolio on day k where t is the announcement date of quarter q’s earnings.

We choose 60 trading days for the post-announcement window because Bernard and

Thomas (1989) report that most of the drift occurs during the first 60 trading days after

the announcement (about three calendar months). Each stock is matched with one of

25 size - B/M portfolios at the end of June based on the market capitalization at the

end of June and B/M, the book equity of the last fiscal year end in the prior calendar

year divided by the market value of equity at the end of December of the prior year.

The daily returns of 25 size - B/M portfolios are from Kenneth French’s website.12

4 The Effect of Competing News on Announcement

Date Returns, Volume, and Post-Earnings An-

nouncement Drift

We specify the investor distraction hypothesis as containing three parts:

12http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data library.html

9

Hypothesis 1a: The sensitivity of the announcement abnormal return to earnings news

decreases with the number of competing announcements.

Hypothesis 1b: The abnormal trading volume on the day of announcement decreases

with the number of competing announcements.

Hypothesis 1c: The sensitivity of the post-announcement abnormal return to earnings

news increases with the number of competing announcements.

Hypotheses 1a and 1c predict opposite directions for the effect of the number of com-

peting news announcements on the announcement-period sensitivity of the firm’s stock

return to its earnings surprise, versus the sensitivity of the post-event return. This helps

distinguish the investor distraction hypothesis from alternative theories. For example, it

might be argued that competing news announcements affect the informativeness of the

firm’s earnings announcement. However, such an alternative theory predicts the same

direction of effect on the announcement period and the post-event sensitivity of returns

to the firm’s earnings surprise (holding constant the fraction of the total response to the

earnings surprise that is delayed).

4.1 Descriptive Statistics

Table 1 reports the descriptive statistics of daily number of quarterly earnings announce-

ments. The mean number of announcements a day is 120.8 and the median number is

71. The percentiles of the number of announcements show that there is a wide varia-

tion in the number of earnings announcements per day; the 10th percentile number of

announcements is 20 and the 90th percentile is 290. Earnings announcements seem to

cluster by day of week and show a highly seasonal pattern. As documented by other

studies, the number of announcements is higher on Tuesday, Wednesday, and Thursday,

and lowest on Friday (e.g., Damodaran (1989), DellaVigna and Pollet (2006)); the aver-

age number of announcements on Friday is 68.8, which is less than a half of the average

number of announcements on Thursday (152.2).

When examined by month, the number of announcements shows an approximately

3-month cycle, with the lowest number of announcements in March, June, September,

and December. This pattern reflects the fact that about 60% of the announcements are

for fiscal quarters ending in March, June, September, and December and that it takes

one to two months from the end of fiscal quarter until the earnings announcement date.13

13In 2002, SEC deadline for filing quarterly reports was 45 days of quarter end and 90 days of yearend for annual reports. The deadlines are now accelerated to 35 days for quarterly reports and 60 daysfor annual reports for fiscal years ending on or after December 15, 2005.

10

Table 2 shows the average size and book-to-market ratios by the decile rank of the

number of announcements. The decile rank is based on quarterly sorts of earnings

announcement observations by the number of announcements on the announcement

day. Earnings announcements on high-news days are from larger firms than those on

low-news days (p < 0.001), while there is no significant difference in book-to-market

ratios between the two extreme number of announcements deciles. However, the size and

book-to-market ratios are not monotonic across the number of announcement deciles.

Deciles 3 and 4 have relatively low size and high book-to-market ratios.

4.2 Announcement Date Returns and Post-Earnings Announce-ment Drift

We first perform univariate analysis to examine the effect of competing news on price

reactions to earnings news. In each calendar quarter, we perform a two-way independent

sort of all quarterly earnings announcements observations in that quarter into 10 ×10 = 100 groups based upon the number of earnings announcements on the day of the

earnings announcement and the earnings surprise (forecast error) as defined in Equation

(1). We exclude earnings announcements for which we lack the size or book-to-market

information of the announcing firm needed to calculate abnormal returns. For each

number of announcements decile, we calculate the mean announcement-period and post-

announcement cumulative abnormal returns for the most positive (FE10) and the most

negative earnings surprise deciles (FE1), and the difference of announcement and post-

announcement cumulative abnormal returns between the two extreme earnings surprise

deciles.

The spread of announcement-day abnormal returns between earnings surprise deciles

10 and 1 (FE10 − FE1) measures the stock price response to earnings news; a larger

spread indicates that investors react more strongly to earnings news on the announce-

ment date. The spread of post-announcement abnormal returns between earnings sur-

prise deciles 10 and 1 measures underreaction to earnings news as reflected in subsequent

drift. If the market is efficient, there will be no difference between good-earnings-news

and bad-earnings-news firms in their post-announcement abnormal returns. A positive

spread indicates underreaction to earnings news – positive abnormal returns following

good news and negative abnormal returns following bad news.

Table 3 shows that investors’ 2-day announcement reactions to earnings news are less

sensitive to earnings news when earnings are announced on high-news days (NRANK=10)

than low-news days (NRANK=1). For the lowest number of announcements decile (low-

news days), the mean spread in 2-day cumulative announcement returns (CAR[0,1])

11

between good earnings news firms (FE10) and bad earnings news firms (FE1) is 7.07%,

whereas for the highest number of announcements decile, the mean spread is 5.67%.

This indicates that the price reactions to earnings news are stronger when earnings are

announced on low-news days than on high-news days.

Greater competing news is also associated with stronger post-earnings announcement

drift. The spread in mean 60-day post announcement abnormal returns (CAR[2,61])

between good and bad earnings news deciles indicates greater underreaction to earn-

ings news on high-news days than on low-news days. For high-news days, the post-

announcement abnormal return spread between extreme earnings surprise deciles is sub-

stantial (7.41%) and highly significant (p < 0.001), whereas the low-news days spread is

smaller (2.81%) and marginally significant(p = 0.055). However, the spread in the post

announcement abnormal returns is not monotonic across the number of announcements

deciles (NRANK). The source of this non-monotonicity may be differences in size and

B/M across the number of announcement deciles (see Table 2). Therefore, we con-

duct multivariate regression analysis in Subsection 5.1 to control for the effect of firm

characteristics such as size, book-to-market, and calendar effects (day of week, month,

or year) on the relation between announcement or post-announcement abnormal returns

and earnings news.

To examine the interaction effect of earnings surprise and the amount of competing

news, we use an ANOVA procedure to test if the abnormal returns spread between high

and low earnings surprise deciles is significantly different between low and high news

days. This is equivalent to testing the significance of the interaction term a3 in the

following regression, using all announcements in the top and bottom of the earnings

surprise deciles and top and bottom of the number of announcements deciles:

CAR = a0 + a1(FE10) + a2(NRANK10) + a3(FE10)(NRANK10) + ε, (3)

where FE10 is an indicator variable that is equal to 1 for the top decile of earnings

surprise, NRANK10 is an indicator variable that is equal to 1 for the top decile of the

number of announcements (high-news days), CAR = CAR[0, 1] for the announcement

date abnormal returns, and CAR = CAR[2, 61] for the post-announcement cumulative

abnormal return. Thus, the ANOVA procedure tests whether CAR spreads between

good and bad earnings news firms are different between high and low news days.

The ANOVA procedure confirms that a greater number of competing announcements

mutes the announcement-date stock price reaction to a firm’s earnings surprise. The last

row of Table 3 reports the difference between high and low news days. The difference

between high and low news days in interdecile spreads of CAR[0, 1] is −1.40% and

statistically significant (p = 0.033). The difference of interdecile spreads of CAR[2, 61]

12

is 4.60%, and also statistically significant (p = 0.034).

Figures 1 and 2 provide graphical evidence that market reactions to earnings news

is weaker on the announcement day and subsequent drifts are stronger when earnings

are announced on high-news days than low-news days. In Figure 1, the abnormal an-

nouncement return (CAR[0,1]) is plotted against earnings surprise deciles, separately

for high-news days (Decile 10) and for low-news days (Decile 1). The market reaction is

less sensitive to earnings news on high-news days, as reflected by its flatter slope in the

graph. Furthermore, it appears that the difference in the sensitivity between high and

low news days is more pronounced among positive rather than negative earnings sur-

prises. We examine the distraction effect separately for positive and negative earnings

surprises later in Subsection 5.1.

Figure 2 shows the mean post-announcement abnormal returns (CAR[2,61]) as a

function of earnings surprise deciles. The slopes of the graphs show that post-announcement

abnormal returns are more predictable based on earnings news when earnings are an-

nounced on high-news days than low-news days, consistent with a more sluggish reaction

to earnings news announced on high-news days. Figures 1 and 2 and the univariate re-

sults in Table 3 suggest that investors react more sluggishly to earnings news when they

are distracted by competing announcements.

5 Regression Analysis

To control for other possible determinants of investor responses to earnings news, we

perform multivariate tests. Subsection 5.1 describes how competing news affects the

sensitivity of returns to a firm’s earnings news. Subsection 5.2 describes how competing

news affects the sensitivity of trading volume to a firm’s earnings news.

5.1 Distraction and the Return Response to Earnings News

In Subsubsection 5.1.1 we describe full sample tests. We perform additional analysis and

consider alternative explanations in Subsubsection 5.1.2. Then in Subsubsection 5.1.3

we provide tests in subsamples of positive versus negative earnings surprises.

5.1.1 Full Sample Tests

In order to control for possible sources of variation in the relation between announcement

date returns and earnings news and also between post-announcement drifts and earnings

news, we run regressions of two-day announcement abnormal return (CAR[0,1]) or 60-

day post-announcement abnormal return (CAR[2,61]) on the earnings surprise decile

13

rank (FE), the number of announcements decile rank (NRANK), the interaction term

(FE×NRANK), control variables, also interacted with FE,

CAR = a0+a1FE+a2NRANK+a3(FE×NRANK)+n∑i=1

ciXi+n∑i=1

bi(FE×Xi)+ε. (4)

The investor distraction hypothesis posits that announcement return is less sensitive and

post-announcement return is more sensitive to earnings news on high news days. Thus,

we expect a3 < 0 when we use CAR[0,1] as dependent variable and a3 > 0 when we use

CAR[2,61] as dependent variable.

Following past literature, we use the decile rank of forecast error as opposed to the

forecast error itself. This reduces the influence of outliers, and linearizes the relation

between abnormal returns and the earnings surprise (see Figure 1). Previous research

shows that investor reactions to earnings news vary with firm size, book-to-market ra-

tios, number of analysts following, reporting lags, institutional ownership, and day of the

week.14 Thus, we include as control variables size and B/M deciles, the number of ana-

lysts following the firm during the most recent fiscal year (Log(1+# analysts following)),

reporting lags (the number of days from the quarter end until the announcement date),

institutional ownership, and day of week/month/year dummies. Institutional ownership

(IO) is the percentage of shares owned by institutions at the end of the most recent

calendar quarter constructed from the CDA/Spectrum 13F database. Standard errors

of regression coefficient estimates are adjusted for heteroskedasticity and clustering by

the day of announcement.

The regression results in Table 4 describe the sensitivity of the immediate stock

price reaction (CAR[0,1]) and the delayed response (CAR[2,61]) to the earnings sur-

prise as a function of the amount of competing news. The coefficient on the interac-

tion between earnings surprise rank and the number of competing announcements rank

(FE×NRANK) measures the effect of competing announcements on market reactions to

earnings news. For the announcement return (CAR[0,1]), the coefficient of the interac-

tion term (FE×NRANK) is negative (−0.021) and significant at the 1% level (Regression

1). Since the coefficient estimate for FE is 1.027, this implies that the market reactions

are about 20% less sensitive to earnings news on high-news days (NRANK=10) com-

pared to those on low-news days (NRANK=1).15

For post-announcement abnormal returns (CAR[2,61]), the coefficient on the inter-

action between earnings surprise decile rank and the number of announcements rank

(FE×NRANK) is positive (0.049) and significant at the 1% level after controlling for

14E.g., Bernard and Thomas (1989), Chambers and Penman (1984), DellaVigna and Pollet (2006).15The sensitivity is 1.027− (0.021× 10) = 0.817 for NRANK=10 and 1.027− (0.021× 1) = 1.006 for

NRANK=1.

14

the effect of size, B/M, analyst following, reporting lags, institutional ownership, and

day of week, month, and year (Regression 5).

Thus, consistent with the distraction hypothesis, distracting news has opposite effects

on the sensitivity of a firm’s announcement period returns to its earnings surprise versus

the sensitivity of its post-announcement reaction to its earnings surprise. These opposite

effects raise a challenging hurdle to alternatives explanations such as one based upon

the informativeness of earnings announcements.

We also examine whether the distraction effect varies with firm size, analyst follow-

ing, or institutional ownership. The distraction effect is measured by the coefficient esti-

mate of FE×NRANK, which is negative in the announcement-return regression (weaker

response for high-news day announcements) and positive in the post-announcement re-

gression (stronger drift after high-news day announcements). To test whether firm size,

analyst following, or institutional ownership affects the distraction effect, we include a

triple interaction term (FE×NRANK×Xi, where Xi=SIZE, Log(1+#Analysts), or IO)

in Regressions (2)-(4) and (6)-(8). There is weak evidence that the distraction effect is

stronger among firms with small market capitalizations, low analyst following, or low

institutional ownership. For example, FE×NRANK×SIZE is positive and significant

at 10% in Regression (2), implying that the distraction effect on announcement re-

turns (negative FE×NRANK) is muted among larger firms. For the post-announcement

drift, FE×NRANK×SIZE is negative and significant at 10% in Regression (6), implying

that the distraction effect on post-announcement returns (positive FE×NRANK) weaker

among larger firms.

5.1.2 Additional Analysis and Alternative Explanations

We consider several alternative explanations of our findings. First is the presence of

errors in announcement dates. The sensitivity of the announcement-day abnormal return

to earnings news can be weaker and the sensitivity of the post-announcement abnormal

return to earnings news stronger on high-news days if there are more errors in our

announcement dates in the top number of announcement decile (high-news days). If the

actual announcement date is later than our announcement date, our post-announcement

period [2,61] may include the actual announcement date. A higher incidence of such

errors among high news days could contribute to the weaker/stronger relation between

announcement/post-announcement abnormal returns and earnings surprises on high-

news days.

We check the accuracy of our announcement dates with the news-wire announcement

dates of DellaVigna and Pollet (2006).16 DellaVigna and Pollet (2006) hand-collected

16We thank Stefano DellaVigna for suggesting this robustness check and Stefano DellaVigna and

15

2,766 randomly-selected earnings announcements from 1984 to 2003 using Lexis-Nexis

and the PR news-wires. We are able to match 159 out of 11,764 low-news day an-

nouncements (NRANK=1) with the news-wire announcement dates. Out of 159 an-

nouncements, 3 announcements are different from the news wire announcement dates;

two of our announcement dates are earlier than the news wire dates by 1 day, one

is earlier than the news wire date by 3 or more days. For high-news day announce-

ments (NRANK=10), 59 out of 12,185 announcements are matched with the news-wire

announcement dates and 3 of the matched announcements are different from the news-

wire dates; for all three, our dates are earlier than the news wire dates by one day. The

error rate is indeed greater in the high news days than low news days. However, the

sample is too small for the difference in error rates to be statistically significant.

To address the possible effect of errors in announcement dates on return sensitiv-

ities, we compute announcement and post-announcement abnormal returns using dif-

ferent windows. The dependent variable in Regression 1 in Table 5 is the abnormal

return over days [−1,1] of the announcement. Including day −1 can help in case the

actual announcement as reported in news wires is earlier than our announcement date

imputed from Compustat and IBES.17 When the actual announcement is later than our

announcement date, our post-announcement window may capture the actual announce-

ment date and therefore we could find stronger predictability of post-announcement

abnormal returns based on earnings news.18 To address this possibility, in Regression 2

in Table 5, we start the post-announcement window from day 3. We find that the results

are quite similar when we use different windows for computing announcement and post-

announcement abnormal returns. We therefore conclude that errors in announcement

dates are not likely to affect the results.

Second, it is possible that high-news days are associated with weaker reactions on the

announcement date because there is more leakage of earnings news before the announce-

ment. To address this possibility, in Regression 3 we examine 30-day abnormal returns

before the announcement. We find that in this pre-announcement period, the relation

between announcement abnormal returns and earnings news does not differ significantly

across NRANK. Thus, there is no indication of any difference in information leakage in

relation to NRANK.

One may argue that the number of distracting events affects the informativeness

of the firm’s earnings about fundamental value. We do not see any clear reason why

Joshua Pollet for providing the data.17However, we did not find any such cases among our announcement dates matched with the news-wire

dates from DellaVigna and Pollet (2006).18However, none of the actual (news-wire) announcement date falls in the post-announcement window

[2,61] for high-news day announcements in our matched announcement dates sample.

16

the number of competing announcements should affect the informativeness of the given

firm’s earnings surprise or the sensitivity of its stock price to its own earnings surprise,

or any presumption as to which way such an effect would go. However, if a given firm’s

earnings were more informative at times when there are few competing announcements,

we would expect both the immediate and total price response to the firm’s earnings

announcement to be larger at such times. To evaluate this alternative hypothesis, we

test directly whether low-news days are associated with a stronger total sensitivity of

stock prices to earnings news. Regression 4 shows that the sensitivity of abnormal

returns over days [0,61] to earnings news, which measures the total impact of earnings

news on stock prices, does not differ across the number of announcement deciles.

Another alternative interpretation of our findings is that in the short run the amount

of capital available to sophisticated investors to arbitrage post-earnings announcement

drift is limited. Since in practice even short-selling requires capital, when NRANK is

high there are many competing uses for arbitrage capital (to exploit the large number

of earnings surprises). So even if there is no limited attention, high NRANK could

imply a lower immediate reaction to a given earnings surprise, and therefore stronger

post-announcement drift.

However, the arbitrage capital argument is actually ambiguous, because earnings

announcements can free up, rather than take up, arbitrage capital. If arbitrageurs have

already taken a position in a stock for any reason, then when it announces earnings and

uncertainty is resolve, arbitrageurs have an opportunity to reverse out their positions.

Furthermore, even if it be the case that arbitrageurs are on the whole committing

more rather than less capital in response to earnings surprises, the capital constraints

argument has a counterfactual implication. Professional arbitrageurs are most active in

large and more liquid stocks. This suggests that small naive investors would be more im-

portant for the price-setting of small stocks, and sophisticated professional arbitrageurs

would have the greatest influence on the prices of large and liquid stocks. If so, then

shocks to the scarcity of arbitrage capital will have a bigger effect on arbitrage activity

in larger stocks. Thus, the arbitrage capital argument implies that the distraction effect

should be greatest among larger stocks. However, our evidence in Table 4 indicates if

anything the opposite–the distraction effect seems stronger among small stocks.

5.1.3 Positive versus Negative Earnings Surprises

Past research has shown that stock returns are more sensitive to the size of positive

earnings surprises than the size of negative ones (Hayn (1995)). Since these reactions

are asymmetric, it is interesting to examine separately the effect of competing announce-

ments for positive and negative earnings surprises.

17

Table 6 reports regression evidence for the positive and negative earnings surprise

subsamples. NRANK is the decile rank of the number of announcements as defined

earlier and FE is the quintile rank of the earnings surprise within each subsample of

positive or negative earnings surprises.

The effect of competing news on market reactions to earnings news is very different

in the two samples. As with the full sample, for positive earnings news, the number of

announcements significantly reduces the sensitivity of announcement abnormal return to

positive earnings news. Similarly, the number of announcements significantly increases

the sensitivity of post-announcement abnormal return to positive earnings news. For the

announcement return regression, the coefficient of the interaction term FE×NRANK is

−0.039 and significant at the 1% level, and in the post-announcement return regression

is 0.120 and significant at the 5% level.

In contrast, in the negative earnings news subsample, the effect of competing news

on the market reactions to news is insignificant for the immediate reaction, though it is

positive (0.156) and significant for long-term drift. While the coefficient of the interaction

term FE ×NRANK is greater for the negative surprise sample (0.156 versus 0.120), it

seems that the significance of the term FE×NRANK for the negative surprise sample is

driven not by underreaction, but by the relation between NRANK and the sensitivity of

the total price reaction to earnings news (in the regression with CAR[0,61] as dependent

variable, FE × NRANK is significant at the 5% level for the negative surprise sample

and insignificant for the positive surprise sample). Therefore, the results suggest that

distraction has a greater effect on underreaction to positive than to negative earnings.

A possible explanation lies in the fact that attention-drawing events, including ex-

treme news, are on average associated with individual investor purchases (Barber and

Odean (2006)). Barber and Odean suggest that this results from the combination of

limited attention, and the asymmetry between stock buying and selling. An investor

selects stocks to purchase from a potential universe of thousands of stocks, but sells are

mostly limited to the few stocks in the investor’s portfolio. Thus, extreme news about

a stock, whether good or bad, brings investor attention to that stock, and on average

leads to a greater preponderance of buying over selling.19

This suggests that attention-driven purchases after bad news may weaken market

reactions to this news. To the extent that distraction mutes attention-driven contrarian

trading after negative news, distraction can strengthen instead of weaken the immediate

response to the negative surprise. This effect can potentially offset the basic distraction

19Although their proxies for investor attention do not include distracting information, Hou, Peng, andXiong (2006) also provide evidence suggesting that attention has a larger effect on market reactionsto positive than to negative earnings news, and refer to the asymmetry between buying and sellingdecisions.

18

effect that motivates our tests.

The reasoning is different after good news, which, as shown by Barber and Odean, is

also associated with attention-driven buys. Such buying should tend to magnify the price

reaction to the positive news. To the extent that distraction mutes this attention-driven

buying after positive news, distraction will tend to weaken the immediate response to

the positive surprise. Thus, this effect potentially reinforces the basic distraction effect

that motivates our tests.

In summary, when we take into account that the effect of distraction is likely to be

different for investor buys and sells, there are possible offsetting effects on how distraction

affects price reactions to bad news; but the prediction remains unambiguous for the effect

of distraction on price reactions to good news.

5.2 Distraction and the Volume Response to Earnings News

The extent to which investors react to earnings news can also be measured by trading

volume in response to the earnings announcement. The investor distraction hypothesis

holds that competing announcements will mute the trading volume response to earnings

news.

We define abnormal volume on day j relative to the announcement date t as a

normalized difference between the log dollar volume on day j and the average log dollar

volume over days [−41,−11] of the announcement:

V OL[j] = Log (DollarV olt+j + 1)− 1

30

t−11∑

k=t−41

Log(DollarV olk + 1). (5)

We perform regression analysis of the abnormal trading volume on the announce-

ment day (VOL[0]) and also two days around announcement (VOL[0,1]). The two-day

abnormal trading volume is defined as the average of abnormal trading volumes on the

announcement date (VOL[0]) and on the following day (VOL[1]).

Since both extreme positive and negative earnings surprises are likely to generate

large trading volume, we regress the one- or two-day abnormal trading volume on the

decile rank of absolute earnings surprises (AFE), the number of announcement decile

rank (NRANK), and other control variables. In addition to the size, B/M, number of

analysts following, reporting lags, institutional ownership, and year/month/day of week

dummies we employed in the previous section, we also include market abnormal trading

volume during the same period (MKTVOL for one-day, MKTVOL2 for two-day window)

so that we are not capturing the market-wide variations in trading volumes. MKTVOL

19

is the average abnormal volume of all CRSP firms on that day where the abnormal

volume of each firm is defined in a manner similar to (5).

Table 7 shows that the coefficient of NRANK is negative and significant at the

1% level in all regression models, indicating that both 1- and 2-day abnormal volume is

lower when earnings are announced on high-news days compared to low-news days, after

controlling for the effect of earnings news or size of earnings news, and market trading

volume, firm-characteristics, and calendar effects. To control for a possible non-linear

effect of earnings surprise on trading volume, we also use indicator variables for each

earnings surprise decile instead of the absolute earnings surprise decile rank (Regressions

3 and 6) and find similar results. Overall, the results show that investor reaction to

earnings news as measured by abnormal trading volume is weaker when earnings are

announced on high-news days.

6 Portfolio Trading Strategies

We now test whether investors can use the distraction effect to form better portfolios.

Based on the previous results, we expect investors to be able to achieve superior returns

by combining earnings surprise information with information about distracting news, as

measured by the number of competing earnings announcements.

At the end of each month from March 1995 until December 2004, we independently

sort stocks into 5×5 portfolios based on their most recent earnings surprises within

the last three months and the number earnings announcements on the day of earnings

announcement. We then calculate equally-weighted returns of each of the resulting 5×5

portfolios during the following month. Within each number of announcements quintile

(NRANK), we form a hedge portfolio that is long in the good news portfolio (FE =

5) and short in the bad news portfolio (FE = 1) and compute the return of the hedge

portfolio.

If investors underreact to earnings news, the good news portfolio will outperform

the bad news portfolio. Therefore, the abnormal return of the hedge portfolio will

be larger when there is stronger post-earnings announcement drift. Since quarterly

earnings announcements during the preceding three months are used to form portfolios,

the strategy uses most of CRSP stocks with quarterly earnings information and captures

up to three months of post-announcement returns. We measure monthly abnormal

performance of the portfolio using alphas from a time-series regression of the portfolio

return (less the risk-free rate except for the zero-cost hedge portfolios) on Fama-French

three factors.

Table 8 shows that a trading strategy designed to exploit post-earnings announce-

20

ment drift achieves a higher abnormal return when implemented on earnings announce-

ments that occur on high-news days than on low-news days. The Good minus Bad

earnings news hedge portfolio 3-factor alpha is 1.64% per month and significant at the

1% level for the high-news portfolio (NRANK=5). The alpha is only 0.77% and insignif-

icant for the low-news portfolio (NRANK=1). In other words, there is no discernible

post-earnings announcement drift when there is little competing news that distracts in-

vestors from the earnings news. The hedge portfolio returns across NRANK are not

monotonic, possibly due to small firm sizes in the second number of announcement

quintile (deciles 3 and 4 in Table 2).

The “fund-of-fund” portfolio, which is formed by going long in the high-news hedge

portfolio and short the low-news hedge portfolio, has a 3-factor alpha of 0.86% which

is significant at the 10% level. This indicates that the post-announcement drift portfo-

lio strategy using high-news day announcements earns marginally higher returns than

that using low-news day announcements. Also the evidence suggests that the higher

profitability of high-news portfolio is mainly due to greater distraction effect on positive

news. The 3-factor alpha of the high-minus-low portfolio (NRANK5-NRANK1; last row

of Table 8) which is long in high-news portfolio and short in low-news portfolio is 0.48%

(significant at 10% level) for the top earnings surprise quintile (FE=5) while it is −0.38%

and insignificant for the bottom earnings surprise quintile (FE=1).

The portfolio strategy findings confirm the univariate and regression results that post

earnings announcement drift is stronger for earnings announcements made on high-news

days than low-news days. The portfolio findings also indicate that, in the absence of

transactions costs, the amount of distraction on the day of the earnings announcement

is useful information for an investor who seeks to exploit post-earnings announcement

drift to achieve superior returns.

7 Related versus Unrelated Announcements

So far we have treated all announcements by other firms alike. However, it is possible

that not all competing announcements are distracting. If two firms are closely related,

an announcement by a one firm might attract attention to the other. For example,

Google’s announcement of its earnings may attract investors’ attention to earnings an-

nouncements by other internet portal firms, and distract investor attention from the

earnings announcements of totally unrelated firms. On the other hand, announcements

by related firms can be more distracting to investors who specialize in a particular sector.

Therefore, it will be interesting to examine how related and unrelated announcements

affect market reactions to earnings news differently.

21

We identify firm relatedness by whether they are in the same industry using the

Fama-French 10 industry classification. The number of related announcements is the

number of earnings announcements by same-industry firms; the number of unrelated

announcements is the number of announcements by firms in other industries. Since

Industry 10 in the Fama-French classification is defined as ‘others’ which do not belong

to any of the pre-defined categories, we limit the analysis to Industries 1-9.

In the regression analyses, we interact the earnings surprise decile rank (FE) deciles

with the number of related announcements decile (NrelRank) or the number of unre-

lated announcements decile (NunrelRank) to test the effect of related and unrelated

announcements on investor reactions to earnings news. To address the possibility that

some industries on average may have a larger number of related or unrelated announce-

ments, the number of related/unrelated announcements deciles (NrelRank/NunrelRank)

are based on quarterly sorts by the number of related/unrelated announcements within

each industry.20

Table 9 shows that unrelated news distracts investors more strongly than related

news. The number of related or unrelated announcements lowers the sensitivity of an-

nouncement returns to earnings news when we examine them separately; the coefficient

of the interaction term of earnings surprise and the number of related announcement

deciles (FE×NrelRank) is −0.015 and significant at the 1% level in Regression 1, and

the coefficient of the interaction term of earnings surprise and the number of unre-

lated announcement deciles (FE×NunrelRank) is −0.021 and significant at the 1% level

in Regression 2. When we include both variables, the effect of unrelated news domi-

nates that of related news; FE×NunrelRank is negative significant at the 5% level and

FE×NrelRank is insignificant in Regression 3.

We also examine the effect of related and unrelated announcements on post-earnings

announcement drift. We find complementary results; the number of unrelated announce-

ments increases the sensitivity of post-announcement returns to earnings news whether

or not related announcements are included, but the effect of related news is not signif-

icant. The coefficient estimates for the interaction term FE×NunrelRank are positive

and significant at the 5% level in Regressions 5 and 6, while the coefficient estimates for

the interaction term FE×NrelRank are insignificant in Regressions 4 and 6. In summary,

Table 9 provides evidence that it is primarily unrelated news firms that mutes the initial

market reaction to earnings news, and thereby intensifies post-earnings announcement

drift.

20The results are similar when we form the number of related/unrelated announcements deciles usingall sample firms and control for possible industry effects by including industry dummies.

22

8 Why Distraction Can Affect Security Prices

A limited attention explanation for an asset pricing pattern must explain why investors

who are not paying full attention would participate in markets and affect prices, and why

any such effects are not eliminated by the trades of fully rational arbitrageurs. The issue

of arbitrage has been addressed by many papers in behavioral finance; if the risk-bearing

capacity of fully rational individuals is finite, then their beliefs do not dominate prices

in the short run; instead, prices reflect a weighted average of investor beliefs, where the

weights depend on the frequencies of different investor types in the population and on

their risk tolerance.21

In the long run, we might expect wealth on average to flow from less rational traders

to more rational traders, which could diminish the influence of imperfectly rational

traders on prices. Again, this is a standard issue in the behavioral finance literature,

and there are some standard responses: that stock prices are noisy so that this wealth-

transfer process can be slow, that in the long run new generations of naive traders

enter the market, and that owing either to aging or to psychological biases in learning

processes some investors may learn to be less rather than more rational over time.

In the specific context of limited attention, owing to cognitive resource constraints,

all investors have limited attention, so there is no way for a flow of wealth to fully

eliminate its effects (see, e.g., Hirshleifer and Teoh (2005)). Even if some investors

allocate resources heavily to the study of a given stock at a given time, and therefore are

highly attentive toward that stock, this entails withdrawal of cognitive resources from

other activities, so we cannot conclude that wealth will tend to flow toward such an

investor.

A different argument set of arguments against limited attention affecting security

pricing are based on investors being able to adjust intelligently to deal with limited

attention. For example, one can leverage attention by focusing on more important

signals. However, it can be hard to know how important an item is until it is carefully

processed.

Alternatively, an investor who is neglecting relevant public information about a stock

could withdraw from trading in that stock, so that his beliefs are not reflected in its

price. However, the same processing and memory constraints that cause neglect of a

signal also make it hard to compensate optimally for the failure to attend to it. For

example, an investor whose valuation disagrees with the market price may inattentively

fail to reason through why the market price differs from his own valuation. Experimental

evidence that the presentation format of decision problems affects choice (e.g., Tversky

21In the context of reactions to earnings news, see the model of Hirshleifer and Teoh (2005).

23

and Kahneman (1981)) indicates that individuals do not compensate optimally for the

limitations in their information processing. Empirically, there are strong indications that

investors are very willing to trade even when they do not possess superior valuations

(Barber and Odean (2000)).22

Other psychological evidence also indicates that individuals do not fully compen-

sate for the fact that they do not possess all relevant information. For example, when

presented with one-sided arguments and evidence and asked to judge a legal dispute,

experimental subjects were biased in favor of the side they heard (Brenner, Koehler,

and Tversky (1996)). As the authors state, “The results indicate that people do not

compensate sufficiently for missing information even when it is painfully obvious that

the information available to them is incomplete.”

Furthermore, we argue that it is reasonable to trade even though one is neglecting

some information. Traditional models of information and securities markets such as

Grossman and Stiglitz (1976) provide the insight that, owing to liquidity or noise trading,

prices aggregate information imperfectly. In consequence, these models imply that even

an uninformed individual who is trading against others who are better informed should

trade based upon his beliefs rather than lapsing into passivity. Intuitively, such an

investor benefits by supplying liquidity to the market, and taking advantage of any

mispricing created by liquidity trades. Analogously, even an investor who neglects a

public signal can benefit from contrarian trading (e.g., through limit orders) based upon

his beliefs. Such contrarian trading could on average be profitable23, yet could also

induce price underreaction to public news events such as earnings announcements.

9 Concluding Remarks

This paper contributes to a growing literature indicating that limited investor atten-

tion has important effects. A distinctive aspect of our tests is a specific focus on the

extraneous news that is the presumptive source of possible investor neglect of relevant

news. We propose the investor distraction hypothesis, which holds that the arrival of

extraneous earnings news causes trading volume and market prices to react sluggishly

to relevant news about a firm. Specifically, we test the investor distraction hypothesis

22An additional possible source of such neglect is overconfidence, a well-documented psychologicalbias. An overconfident individual who wrongly thinks that he has already incorporated the most theimportant signals may not perceive the urgency of adjusting for the fact that he is neglecting a relevantpublic signal.

23Kaniel, Saar, and Titman (2004) find that contrarian trading allows U.S. individual investors toearn positive excess returns in the month after their trades; Linnainmaa (2003) finds that individualday traders in Finland provide liquidity to the market through limit orders and on average profit duringthe day by doing so.

24

by examining how the number of earnings announcements by other firms affects the sen-

sitivity of a firm’s volume, announcement-period return, and post-event return reaction

to its earnings surprise.

Our evidence indicates that the presence of a large number of competing earnings

announcements by other firms is associated with a weaker announcement-date price

reaction to a firm’s own earnings surprise, a lower volume reaction, and stronger sub-

sequent post-earnings announcement drift. A portfolio trading strategy that exploits

post-earnings announcement drift achieves superior performance when implemented on

earnings announcements on days with a large number of competing announcements than

those on days with little competing news. Competing announcements made by firms in

other industries have a stronger distraction effect, whereas those by same-industry firms

do not have a significant effect.

These findings generally support the investor distraction hypothesis. Furthermore,

they indirectly suggest that investors’ limited attention may drive the basic anomaly,

post-earnings announcement drift. More broadly, this evidence raises the possibility that

limited attention is the source of the general pattern documented in several studies of

underreaction to a variety of public corporate news events.

25

References

Baddeley, A., 1990, Human Memory: Theory and Practice (1st ed.). (Lawrence Erl-baum Associates Ltd, London, Erlbaum).

Bagnoli, M., M. B. Clement, and S. G. Watts, 2005, Around-the-clock media coverageand the timing of earnings announcements, Purdue University Working paper.

Baker, A. G., P. Mercier, F. Valleettourangeau, R. Frank, and M. Pan, 1993, Selectiveassociations and causality judgments – Presence of a strong causal factor may re-duce judgements of a weaker one, Journal of Experimental Psychology – Learning,Memory and Cognition 19, 414–432.

Barber, B. and T. Odean, 2000, Trading is hazardous to your wealth: The commonstock investment performance of individual investors, Journal of Finance 55, 773–806.

Barber, B. and T. Odean, 2006, All that glitters: The effect of attention and news onthe buying behavior of individual and institutional investors, Review of FinancialStudies, forthcoming.

Bernard, V. L. and J. K. Thomas, 1989, Post-earnings-announcement drift: Delayedprice response or risk premium?, Journal of Accounting Research, Supplement 27,1–48.

Brennan, M. J., N. Jegadeesh, and B. Swaminathan, 1993, Investment analysis and theadjustment of stock prices to common information, Review of Financial Studies 6,799–824.

Brenner, L. A., D. J. Koehler, and A. Tversky, 1996, On the evaluation of one-sidedevidence, Journal of Behavioral Decision Making 9, 59–70.

Broadbent, D. E., 1958, Perception and communication (Pergamon Press, New York,NY).

Busemeyer, J. R., I. J. Myung, and M. McDaniel, 1993, Cue competition effects –empirical tests of adaptive network learning models, Psychological Science 4, 190–195.

Chambers, A. E. and S. H. Penman, 1984, Timeliness of reporting and the stock pricereaction to earnings announcements, Journal of Accounting Research 22, 21–47.

Cherry, E. C., 1953, Some experiments on the recognition of speech, with one and twoears., Journal of the Acoustical Society of America, 25, 975–979.

Cohen, L. and A. Frazzini, 2006, Economic links and predictable returns, workingpaper, Yale School of Management and University of Chicago.

Damodaran, A., 1989, The weekend effect in information: A study of earnings anddividend announcements?, Review of Financial Studies 4, 607–623.

DellaVigna, S. and J. Pollet, 2006, Investor inattention, firm reaction, and Fridayearnings announcements, UC Berkeley and UIUC, working paper.

DellaVigna, S. and J. M. Pollet, 2007, Demographics and industry returns, UC Berke-ley and UIUC, working paper.

26

Francis, J., D. Pagach, and J. Stephan, 1992, The stock market response to earn-ings announcements released during trading versus nontrading periods, Journal ofAccounting Research 30, 165–184.

Gabaix, X. and D. Laibson, 2004, Bounded rationality and directed cognition, MITand Harvard University, working paper.

Gabaix, X., D. Laibson, G. Moloche, and S. Weinberg, 2006, Information acquisition:Experimental analysis of a boundedly rational model, American Economic Review .forthcoming.

Grossman, S. J. and J. E. Stiglitz, 1976, Information and competitive price systems,American Economic Review 66, 246–253.

Hayn, C., 1995, The information content of losses, Journal of Accounting and Eco-nomics 20, 125–153.

Hirshleifer, D., 2001, Investor psychology and asset pricing, Journal of Finance 64,1533–1597.

Hirshleifer, D., S. S. Lim, and S. H. Teoh, 2004, Disclosure to an audience with limitedattention, Fisher College of Business, Ohio State University.

Hirshleifer, D., J. Myers, L. A. Myers, and S. H. Teoh, 2003, Do individual investorsdrive post-earnings announcement drift?, Fisher College of Business, Ohio StateUniversity, and University of Illinois at Urbana-Champaign.

Hirshleifer, D. and S. H. Teoh, 2003, Limited attention, information disclosure, andfinancial reporting, Journal of Accounting & Economics 36, 337–386.

Hirshleifer, D. and S. H. Teoh, 2005, Limited investor attention and stock marketmisreactions to accounting information, Working paper, Ohio State University.

Ho, T. and R. Michaely, 1988, Information quality and market efficiency, Journal ofFinancial and Quantitative Analysis 5, 357–386.

Hong, H., T. Lim, and J. Stein, 2000, Bad news travels slowly: Size, analyst coverageand the profitability of momentum strategies, Journal of Finance 55, 265–295.

Hong, H., W. Torous, and R. Valkanov, 2007, Do industries lead stock markets?,Journal of Financial Economics .

Hou, K., 2007, Industry information diffusion and the lead-lag effect in stock returns,Review of Financial Studies, forthcoming.

Hou, K. and T. Moskowitz, 2005, Market frictions, price delay, and the cross-sectionof expected returns, Review of Financial Studies , 981–1020.

Hou, K., L. Peng, and W. Xiong, 2006, A tale of two anomalies: The implication ofinvestor attention for price and earnings momentum, Working Paper, Ohio StateUniversity.

Huberman, G. and T. Regev, 2001, Contagious speculation and a cure for cancer,Journal of Finance 56, 387–396.

Ikenberry, D., J. Lakonishok, and T. Vermaelen, 1995, Market underreaction to openmarket share repurchases, Journal of Financial Economics 39, 181–208.

Kahneman, D., 1973, Attention and Effort (Prentice-Hall, Englewood Cliffs, NewJersey).

27

Kaniel, R., G. Saar, and S. Titman, 2004, Individual investor trading and stockreturns, Journal of Finance.

Klibanoff, P., O. Lamont, and T. A. Wizman, 1998, Investor reaction to salient newsin closed-end country funds, Journal of Finance 53, 673–699.

Kruschke, J. K. and M. K. Johansen, 1999, A model of probabilistic category learning,Journal of Experimental Psychology: Learning, Memory, and Cognition 25, 1083–1119.

Linnainmaa, J., 2003, The anatomy of day traders, AFA 2004 San Diego Meetings.

Linnainmaa, J., 2007, The limit order effect, University of Chicago Working Paper.

Loughran, T. and J. Ritter, 1995, The new issues puzzle, Journal of Finance 50,23–52.

McLeod, P., 1977, Parallel processing and the psychological refractory period, ActaPsychologica 41, 381–391.

Moray, N., 1959, Attention in dichotic listening: Affective cues and the influence ofinstructions., Quarterly Journal of Experimental Psychology 11, 56–60.

Pashler, H. and J. C. Johnston, 1998,. Attentional limitations in dual-task perfor-mance, . In H. Pashler, ed., Attention, pp. 155–189. (Psychology Press/Erlbaum(Uk) Taylor & Francis, Hove, England UK).

Pashler, H. E., 1998, The Psychology of Attention (MIT Press, Cambridge, MA).

Peng, L., 2005, Learning with information capacity constraints, Journal of Financialand Quantitative Analysis, forthcoming.

Peng, L. and W. Xiong, 2006, Investor attention, overconfidence and category learning,Journal of Financial Economics, forthcoming.

Peng, L., W. Xiong, and T. Bollerslev, 2006, Investor attention and time-varyingcomovements, European Financial Management .

Pollet, J. M., 2005, Predicting asset returns with expected oil price changes, Universityof Illinois at Urbana-Champaign Working Paper.

Riley, D. and H. L. Roitblat, 1978,. Selective attention and related cognitive processesin pigeons, . In S. H. Hulse, H. Fowler, and W. K. Honig, eds., Cognitive Processesin Animal Behavior, pp. 249–276. (Erlbaum, Hillsdale, NJ).

Seasholes, M. S. and G. Wu, 2005, Predictable behavior, profits, and attention, UCBerkeley Working Paper.

Simons, D. and C. Chabris, 1999, Gorillas in our midst: sustained inattentional blind-ness for dynamic events, Perception 28, 1059–1074.

Simons, D. J. and D. T. Levin, 1997, Change blindness, Trends in Cognitive Sci-ences 1, 261–267.

Sims, C., 2003, Implications of rational inattention, Journal of Monetary Eco-nomics 50, 665–690.

Stroop, J., 1935, Studies of interference in serial verbal reactions, Journal of Experi-mental Psychology 28, 643–662.

28

Treisman, A. and A. Davies, 1973,. Dividing attention to ear and eye, . In S. Kornblum,ed., Attention and Performance IV, pp. 101–117. (Academic Press, New York).

Tversky, A. and D. Kahneman, 1981, The framing of decisions and the psychology ofchoice, Science 211, 453–458. Jan 30.

29

- 0.04

- 0.03

- 0.02

- 0.01

0

0.01

0.02

0.03

0.04

0.05

1 2 3 4 5 6 7 8 9 10

Earnings Surprise Deciles

CA

R[0

,1]

Low news daysHigh news days

Figure 1. Market Reactions to Earnings News: CAR[0,1]

- 0.04

- 0.03

- 0.02

- 0.01

0

0.01

0.02

0.03

0.04

0.05

1 2 3 4 5 6 7 8 9 10

Earnings Surprise Deciles

CA

R[0

,1]

Low news daysHigh news days

Figure 2. Post-Earnings Announcement Drift: CAR[2,61]

Figure 1 and Figure 2 show the average 2-day announcement cumulative abnormal returns (CAR[0,1]) and 60-day post-announcement cumulative abnormal returns (CAR[2,61]) of quarterly earnings announcements against earnings surprise deciles (1: bad news – 10: good news) for announcements on high-news days (number of announcements decile 10) and those on low-news days (number of announcements decile 1). Earnings surprise and number of announcement deciles are formed based on a quarterly independent double sort of quarterly earnings announcements by the corresponding forecast error and the number of quarterly earnings announcements on the day of announcement.

Table 1. Descriptive Statistics of Daily Number of Earnings Announcements Quarterly earnings announcement dates are from CRSP-Compustat merged database and IBES for the period from January 1995 to December 2004. We use the earnings announcement date from Compustat when the firm is not covered by IBES or when Compustat and IBES dates agree, and use the earlier date when Compustat and IBES announcement dates differ. Weekend earnings announcements are excluded from the sample. The daily number of announcements is the total number of quarterly earnings announcements on each day.

Panel A. Distribution of Daily Number of Announcements

Percentiles Mean Std. Dev P10 P25 Median P75 P90 120.8 129.7 20 33 71 175 290

Panel B. Mean and Median Number of Announcements a Day, by Day of Week/Month/Year

Monday Tuesday Wednesday Thursday Friday Mean 103.7 140.2 138.9 152.2 68.8

Median 62 78 77 90 53

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Mean 116.8 145.4 78.5 195 155.2 32.7 176.5 148.7 34 174.1 154.3 32.4

Median 86 147 68.5 142 123 30 122.5 107 32 128 118.5 31

1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

Mean 100.4 107 112.4 117 121.3 145.5 138.1 130.9 120.4 114.5 Median 62 72 76 66 70 86.5 78 72 66 66

Table 2. Firm Characteristics by the Number of Announcements Deciles In each calendar quarter, we sort quarterly earnings announcements during that quarter with earnings surprise (forecast error), size, and book-to-market (B/M) information into deciles by the number of announcements on the day of the announcement. The number of announcements is the total number of quarterly earnings announcements by all firms covered by CRSP-Compustat on the same day. Table 2 reports the average size and B/M by the number of announcements deciles and the difference of size and B/M between deciles 10 and 1 with p-values. The size and B/M values are calculated at the end of June of each year based on the market value of equity at the end of June and the book value of equity for the last fiscal year end in the previous calendar year divided by the market value of equity for December of the previous calendar year.

Size B/M

Decile1 (low-news days) 2347.1 0.662 2 2916.5 0.799 3 2347.8 0.946 4 2216.8 0.831 5 2718.9 0.756 6 3347.3 0.742 7 3076.7 0.714 8 3284.2 0.664 9 3434.4 0.679

Decile10 (high-news days) 3228.8 0.789 Difference (10-1) 881.7 0.127

p-value <0.001 0.128

Table 3. Cumulative Abnormal Returns of Extreme Earnings Surprise Deciles By Number of Announcements Deciles

Using quarterly earnings announcements from January 1995 to December 2004, we calculate the average 2-day announcement cumulative abnormal returns (CAR[0,1]) and 60-day post-announcement cumulative abnormal returns (CAR[2,61]) for extreme earnings surprise deciles (FE10: good news, FE1: bad news) for each number of announcements deciles (NRANK). Earnings surprise and number of announcement deciles are formed based on quarterly independent double sorts of quarterly earnings announcements by the corresponding forecast error and the number of quarterly earnings announcements on the day of announcement. The significance of the return spread between good and bad news firms (FE10-FE1) is marked by * (significant at 5%) and ** (significant at 1%).

Average CAR[0,1] for Earnings surprise deciles 10 and 1

Average CAR[2,61] for Earnings surprise deciles 10 and 1 NRANK

FE1 FE10 FE10 −FE1 FE1 FE10 FE10−FE1 1 (low-news days) −3.27% 3.81% 7.07%** −0.99% 1.82% 2.81%

2 −3.31% 3.45% 6.76%** −1.67% 3.10% 4.77%**

3 −3.22% 3.52% 6.74%** −2.09% 2.09% 4.18%**

4 −3.54% 2.50% 6.04%** −4.79% 3.25% 8.04%**

5 −3.37% 2.22% 5.59%** −2.19% 2.24% 4.43%**

6 −3.17% 3.16% 6.33%** 0.96% 4.17% 3.21% 7 −3.67% 2.82% 6.49%** −3.32% 2.32% 5.64%**

8 −3.45% 2.79% 6.24%** −2.68% 6.30% 8.98%**

9 −3.49% 2.83% 6.32%** −1.21% 4.16% 5.37%**

10 (high-news days) −3.06% 2.61% 5.67%** −2.97% 4.44% 7.41%**

Difference (10−1) 0.20% −1.20%** −1.40%* −1.98% 2.62% 4.60%*

Table 4. Market Reactions to Earnings News: Regression Analysis

Table 4 reports the multivariate tests of the effects of the number of announcements on the relation between announcement and post-announcement returns and earnings surprises. FE is the earnings surprise deciles (FE=1: lowest, 10: highest) and NRANK is the number of announcement deciles based on quarterly independent sorts by forecast errors and the number of announcements on the day of announcement. SIZE is the size decile and BM is the book-to-market decile based on the most recent June size and book-to-market ratio of the firm using NYSE breakpoints, #Analysts is the number of analysts following the firm during the most recent fiscal year, LAG is the reporting lag defined as the number of days from the quarter end until the announcement date, and IO is the percentage of shares owned by institutions. Also included are year, month, and day of week dummies, and their interaction terms of FE. Standard errors adjusted for heteroskedasticity and clustering by the day of announcement are in parentheses (* significant at 10%; ** significant at 5%; *** significant at 1%).

(1) (2) (3) (4) (5) (6) (7) (8) CAR[0,1] CAR[0,1] CAR[0,1] CAR[0,1] CAR[2,61] CAR[2.61] CAR[2.61] CAR[2.61] FE 1.027*** 1.046*** 1.080*** 1.074*** 1.016*** 0.924*** 0.917*** 0.895***

(0.070) (0.071) (0.073) (0.072) (0.239) (0.244) (0.250) (0.251) NRANK 0.073** 0.068** 0.068** 0.067** −0.297*** −0.274** −0.287*** −0.281***

(0.031) (0.032) (0.031) (0.031) (0.108) (0.108) (0.107) (0.108) FE×NRANK −0.021*** −0.024*** −0.030*** −0.029*** 0.049*** 0.064*** 0.066*** 0.070***

(0.005) (0.005) (0.006) (0.006) (0.018) (0.021) (0.024) (0.024) SIZE 0.267*** 0.267*** 0.267*** 0.267*** −0.012 −0.014 −0.013 −0.014 (0.032) (0.032) (0.032) (0.032) (0.102) (0.102) (0.102) (0.102) FE×SIZE −0.040*** −0.046*** -0.041*** −0.040*** −0.054*** −0.027 −0.053*** −0.053***

(0.005) (0.006) (0.005) (0.005) (0.016) (0.020) (0.016) (0.016) BM 0.250*** 0.250*** 0.250*** 0.250*** −0.174*** −0.173*** −0.174*** −0.174***

(0.020) (0.020) (0.020) (0.020) (0.067) (0.067) (0.067) (0.067) FE×BM −0.034*** −0.034*** -0.034*** -0.034*** 0.015 0.015 0.015 0.015 (0.003) (0.003) (0.003) (0.003) (0.011) (0.011) (0.011) (0.011) LAG 0.022*** 0.022*** 0.022*** 0.022*** 0.020 0.021 0.021 0.021 (0.005) (0.005) (0.005) (0.005) (0.017) (0.017) (0.017) (0.017) FE×LAG −0.007*** −0.007*** −0.007*** −0.007*** −0.008*** −0.008*** −0.008*** −0.007***

(0.001) (0.001) (0.001) (0.001) (0.003) (0.003) (0.003) (0.003) Log(1+#Analysts) −0.437*** −0.437*** −0.436*** −0.437*** 1.496*** 1.496*** 1.494*** 1.497***

(0.109) (0.109) (0.109) (0.109) (0.370) (0.370) (0.370) (0.370) FE×Log(1+#Analysts) 0.051*** 0.051*** 0.023 0.052*** −0.153*** −0.153*** −0.101 −0.154***

(0.018) (0.018) (0.021) (0.018) (0.059) (0.059) (0.067) (0.059) IO −1.255*** −1.255*** -1.256*** -1.254*** −0.200 −0.199 −0.198 −0.203 (0.284) (0.284) (0.284) (0.284) (0.919) (0.919) (0.919) (0.919) FE×IO 0.342*** 0.342*** 0.343*** 0.236*** 0.040 0.041 0.039 0.309 (0.046) (0.046) (0.046) (0.058) (0.146) (0.146) (0.146) (0.203) FE×NRANK×SIZE 0.001* −0.004* (0.001) (0.002) FE×NRANK 0.005*** −0.009 ×Log(1+#Analysts) (0.002) (0.007) FE×NRANK×IO 0.019*** −0.048 (0.006) (0.029) Calendar dummies, interacted with FE

X X X X X X X X

Constant −5.872*** −5.840*** −5.836*** −5.835*** −5.748*** −5.897*** −5.816*** −5.844***

(0.436) (0.437) (0.436) (0.436) (1.535) (1.538) (1.534) (1.535) Observations 117,640 117,640 117,640 117,640 116,414 116,414 116,414 116,414 R-squared 5.6% 5.6% 5.6% 5.6% 0.9% 0.9% 0.9% 0.9%

Table 5. Market Reactions to Earnings News: Additional Analysis

Table 5 reports the multivariate tests of the effects of the number of announcements on the relation between returns and earnings surprises. FE is the earnings surprise deciles (FE=1: lowest, 10: highest) and NRANK is the number of announcement deciles based on quarterly independent sorts by forecast errors and the number of announcements on the day of announcement. Control variables include size deciles and book-to-market (B/M) deciles based on the most recent June size and book-to-market ratio of the firm using NYSE breakpoints, the number of analysts following the firm during the most recent fiscal year (log(1+# analysts following)), reporting lags (the number of days from the quarter end until the announcement date), the percentage of shares owned by institutions (institutional ownership), year, month, and day of week dummies, and interaction terms of FE with all control variables. Standard errors adjusted for heteroskedasticity and clustering by the day of announcement are in parentheses (* significant at 10%; ** significant at 5%; *** significant at 1%).

(1) (2) (3) (4) CAR[-1,1] CAR[3,61] CAR[-30,-1] CAR[0,61] FE 1.154*** 1.100*** 1.796*** 2.117***

(0.075) (0.235) (0.161) (0.255) NRANK 0.057* −0.264** 0.079 −0.192*

(0.033) (0.104) (0.078) (0.115) FE×NRANK −0.021*** 0.046*** −0.017 0.024 (0.005) (0.018) (0.013) (0.020) Controls, interacted with FE X X X X Constant −5.907*** −6.052*** −5.197*** −11.911***

(0.472) (1.500) (0.969) (1.575) # Observations 117,639 116,414 117,639 116,414 R-squared 5.8% 0.9% 2.5% 2.1%

Controls: Size and B/M deciles, log(1+# analysts following), reporting lags, institutional ownership, year, month, and day of week dummies. Interaction terms of FE with all control variables are also included.

Table 6. Positive versus Negative Earnings Surprises

Table 6 reports multivariate tests of the effects of the number of announcements on the relation between returns and earnings surprises, separately for positive and negative earnings surprises. NRANK is the number of announcement deciles and FE is the earnings surprise quintile within each subsample of positive and negative surprises (FE = 1: lowest, 5: highest). Control variables include size deciles and book-to-market (B/M) deciles based on the most recent June size and book-to-market ratio of the firm using NYSE breakpoints, the number of analysts following the firm during the most recent fiscal year (log(1+# analysts following)), reporting lags (the number of days from the quarter end until the announcement date), the percentage of shares owned by institutions (institutional ownership), year, month, and day of week dummies, and interaction terms of FE with all control variables. Standard errors adjusted for heteroskedasticity and clustering by the day of announcement are in parentheses (* significant at 10%; ** significant at 5%; *** significant at 1%).

Positive Earnings Surprises Negative Earnings Surprises CAR[0,1] CAR[2,61] CAR[0,61] CAR[0,1] CAR[2,61] CAR[0,61] FE 1.475*** 0.957 2.582*** 0.538** −1.133 −0.277 (0.178) (0.634) (0.694) (0.257) (0.819) (0.834) NRANK 0.032 −0.345** −0.292** −0.005 −0.593*** −0.526**

(0.039) (0.135) (0.147) (0.067) (0.224) (0.233) FE×NRANK −0.039*** 0.120** 0.074 0.015 0.156*** 0.153**

(0.013) (0.049) (0.055) (0.018) (0.059) (0.061) Controls, interacted with FE X X X X X X

Constant −2.401*** −0.528 −3.173 −5.224*** −1.581 −7.779**

(0.532) (1.921) (2.055) (0.955) (3.146) (3.187) # Observations 64,231 63,643 63,643 36,584 36,064 36,064 R-squared 2.0% 1.1% 1.7% 2.0% 1.6% 1.5%

Controls: Size and B/M deciles, log(1+# analysts following), reporting lags, institutional ownership, year, month, and day of week dummies. Interaction terms of FE with all control variables are also included.

Table 7: Trading Volume Response to Earnings News We perform multivariate analysis of the effect of competing annoucements on trading volume response to earnings news. The dependent variable is one- or two-day abnormal trading volume (VOL[0], VOL[0,1]). Abnormal trading volume on a given day is defined as the log dollar trading volume on that day normalized by the average log dollar trading volume over days [-41,-11] of the announcement. VOL[0] is the abnormal trading volume on the announcement date and VOL[0,1] is the average abnormal trading volume over days [0,1] of the announcement. AFE is the absolute earnings surprise deciles and NRANK is the number of announcements deciles based on quarterly independent sorts by absolute forecast errors and the number of announcements on the day of announcement. Regressions (3) and (6) use indicator variables for each earnings surprise deciles. MKTVOL is the market abnormal trading volume on the announcement date and MKTVOL2 is the 2-day market abnormal trading volume over days [0,1] of the announcement. Other control variables include size and book-to-market (B/M) deciles based on the most recent June size and book-to-market ratio of the firm using NYSE breakpoints the number of analysts following the firm during the most recent fiscal year (log(1+# analysts following)), reporting lags (the number of days from the quarter end until the announcement date), the percentage of shares owned by institutions (institutional ownership), year, month, and day of week dummies. Standard errors adjusted for heteroskedasticity and clustering by the day of announcement are in parentheses (* significant at 10%; ** significant at 5%; *** significant at 1%).

(1) (2) (3) (4) (5) (6) VOL[0] VOL[0] VOL[0] VOL[0,1] VOL[0,1] VOL[0,1] AFE 0.031*** 0.031*** 0.030*** 0.029*** (0.002) (0.002) (0.001) (0.001) NRANK −0.022*** −0.028*** −0.028*** −0.022*** −0.029*** −0.029***

(0.002) (0.003) (0.003) (0.002) (0.002) (0.002) Dummy variables for FE deciles X X

MKTVOL 0.779*** 0.781*** (0.044) (0.043) MKTVOL2 0.818*** 0.820***

(0.046) (0.046) Controls X X X X

Constant 0.470*** 0.634*** 0.686*** 0.530*** 0.718*** 0.794***

(0.016) (0.036) (0.038) (0.014) (0.032) (0.034) # Observations 117,640 117,640 117,640 117,640 117,640 117,640 R-squared 0.7% 3.0% 3.0% 1.0% 4.2% 4.4%

Controls: Size and B/M deciles, log(1+# analysts following), reporting lags, institutional ownership, year, month, and day of week dummies.

Table 8: Fama-French Alphas of Post-Earnings Announcement Drift Portfolios At the end of each month from March 1995 until December 2004, we independently sort stocks into 5x5 groups based on their most recent quarterly earnings surprises within the last three months (FE=1~5) and the number earnings announcements on the day of earnings announcement (NRANK=1~5). We calculate equally-weighted returns of the resulting 5x5 portfolios during the following month. Within each of number of announcements rank (NRANK), we form a hedge portfolio that is long in good news portfolio (FE=5) and short in bad news portfolio (FE=1) to exploit post-earnings announcement drifts. Alphas from time-series regressions of portfolio monthly returns (less the risk-free rate except for the zero-cost hedge portfolios) on Fama-French three factors are reported with Newey-West standard errors with 12 lags are in parentheses (* significant at 10%; ** significant at 5%; *** significant at 1%). The first row reports the alphas of the base post-earnings announcement drift portfolios equally weighting firms in each earnings surprise quintile. The last row reports the alphas of hedge portfolios long in high-news day portfolio (NRANK=5) and short in the low-news day portfolio (NRANK=1) within each earnings surprise quintile.

Earnings Surprise Quintile

FE=1 (Bad news)

2 3 4 FE=5

FE5-FE1 (Good news) (Good −Bad) All −0.57 −0.53*** −0.12 0.04 0.75*** 1.32***

(0.36) (0.20) (0.14) (0.14) (0.26) (0.26) NRANK 1 −0.21% −0.53** −0.17 0.04 0.56** 0.77 (0.42) (0.22) (0.21) (0.21) (0.24) (0.47) 1.55***2 −1.10*** −0.56** −0.04 −0.23 0.45 (0.37) (0.24) (0.21) (0.18) (0.28) (0.26) 3 −0.51 −0.61*** 0.02 −0.02 0.70** 1.22***

(0.43) (0.23) (0.17) (0.15) (0.30) (0.37) 4 −0.16 −0.56** −0.26 0.32 0.94*** 1.09***

(0.50) (0.25) (0.16) (0.20) (0.34) (0.40) NRANK 5 −0.60* −0.44*** −0.15 0.06 1.04*** 1.64***

(0.34) (0.17) (0.14) (0.17) (0.34) (0.31) NRANK5 −0.38 0.09 0.03 0.02 0.48* 0.86*

− NRANK1 (0.33) (0.16) (0.20) (0.21) (0.24) (0.47)

Table 9: The Effect of Related vs. Unrelated Announcements

We examine the effect of the number of related and unrelated announcements on the 2-day announcement cumulative abnormal returns (CAR[0,1]) and the 60-day post-announcement cumulative abnormal returns (CAR[2,61]). We calculate the daily number of quarterly earnings announcements for each industry using Fama-French 10 industry classification. The number of related announcements is the number of quarterly earnings announcements by the same industry firms, and the number of unrelated announcements is the number of quarterly earnings announcements by firms in other industries. FE is the earnings surprise deciles (FE=1: lowest, 10: highest) based on quarterly sorts by forecast errors using all observations, and NrelRank/NunrelRank is the decile rank of the number of related/unrelated announcements (10: highest, 1: lowest) announcements based on quarterly sorts by the number of related/unrelated announcements within each industry. We exclude firms in Industry 10 ('Others'). Control variables include size deciles and book-to-market (B/M) deciles based on the most recent June size and book-to-market ratio of the firm using NYSE breakpoints, the number of analysts following the firm during the most recent fiscal year, reporting lags (the number of days from the quarter end until the announcement date), the percentage of shares owned by institutions (institutional ownership), year, month, and day of week dummies, and interaction terms of FE with all control variables. Standard errors adjusted for heteroskedasticity and clustering by the day of announcement are in parentheses (* significant at 10%; ** significant at 5%; *** significant at 1%).

(1) (2) (3) (4) (5) (6) CAR[0,1] CAR[0,1] CAR[0,1] CAR[2,61] CAR[2,61] CAR[2,61] FE 0.976*** 1.019*** 1.022*** 0.907*** 0.741*** 0.748***

(0.075) (0.079) (0.079) (0.252) (0.271) (0.270) NrelRank 0.063** 0.010 −0.244** -0.044 (0.032) (0.039) (0.105) (0.141)

Controls: Size and B/M deciles, log(1+# analysts following), reporting lags, institutional ownership, year, month, and day of week dummies. Interaction terms of FE with all control variables are also included.

FE×NrelRank −0.015*** −0.004 0.026 -0.011 (0.005) (0.006) (0.017) (0.022) NunrelRank 0.092** 0.084* −0.355*** −0.321*

(0.036) (0.046) (0.123) (0.165) FE×NunrelRank −0.021*** −0.018** 0.052** 0.061**

(0.006) (0.007) (0.021) (0.028) Controls, interacted with FE X X X X X X

Constant −5.941*** −6.154*** −6.163*** −5.765*** −4.939*** −4.921***

(0.470) (0.492) (0.493) (1.658) (1.761) (1.760) Observations 92,685 92,685 92,685 91,737 91,737 91,737 R-squared 5.7% 5.7% 5.7% 1.1% 1.1% 1.1%